SPE 95869

Probabilistic Forecasting for Mature Fields With Significant Production History:

A Nemba Field Case Study
G.R. King, SPE, Cabinda Gulf Oil Co. (Cabgoc); S. Lee, SPE, Chevron Energy Technology Co.; P. Alexandre, SPE,
Soc. Nacional de Combustiveis de Angola (Sonangol); and M. Miguel, SPE, M. Blevens, M. Pillow, and G. Christie,
Cabgoc

Copyright 2005, Society of Petroleum Engineers

and simultaneous, multi-variable input modifications
This paper was prepared for presentation at the 2005 SPE Annual Technical Conference and (experimental design). The sensitivity analysis were used to
Exhibition held in Dallas, Texas, U.S.A., 9 – 12 October 2005.
determine the appropriate uncertainty ranges for individual
This paper was selected for presentation by an SPE Program Committee following review of
information contained in a proposal submitted by the author(s). Contents of the paper, as
uncertain factors while the experimental design was used to
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to determine the interactions of these uncertainty factors.
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at During the experimental design portion of the analysis,
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
modifications were made to the history matched input data and
for commercial purposes without the written consent of the Society of Petroleum Engineers is the simulator was run through both the historical and
prohibited. Permission to reproduce in print is restricted to a proposal of not more than 300
words; illustrations may not be copied. The proposal must contain conspicuous prediction periods of the project. In order to retain the integrity
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.
of the history match, a Quality of History Match (QoHM)
variable was defined to quantify and rank any degradation to
Abstract the history match caused by perturbation of uncertain data.
The Nemba Field is the third largest oilfield in terms of daily Proxy equations for the simulation response were then
production in the Block 0 Concession in the Republic of generated for the QoHM variable and the project result
Angola. Oil production from the field began in January 1996 variables: Estimated Ultimate Recovery (EUR) and discounted
through an Early Production System (EPS). The field reached cumulative oil production. Finally, a two-step Monte Carlo
a peak production rate of approximately 110 MBOPD in simulation approach was used to develop the cumulative
August 2002 and is currently undergoing crestal gas injection distribution functions (CDFs) for all alternatives considered
at an instantaneous Voidage Replacement Ratio (VRR) of for the gas compressor project.
approximately 0.85. The field is being evaluated for the In this two-step Monte Carlo simulation, the uncertain
installation of a new compression train to allow for gas factors were sampled from their individual CDFs for use in a
injection at or above a VRR of 1.0. Monte Carlo trial. The sampled data were first input into the
To support this effort, along with other reservoir proxy equation for the QoHM variable to determine the impact
management activities, a reservoir characterization and on the History Match. If the selected input data were found to
simulation study was performed in 2003/2004. The coarse- meet predetermined acceptance criteria for the QoHM
grid, reservoir simulation model generated from this study was variable, then they were passed to the second step: input to the
a six component equation-of-state (EOS) model containing proxy equations for the project result variables. Input data
approximately 246,906 total cells (136,675 active) with combinations that did not meet the acceptance criteria were
approximately 33,794 non-neighbor connections to account discarded. This process was repeated for the desired number of
for complex faulting and truncation across stratigraphic Monte Carlo trials to complete the simulation.
intervals. The workflow for this study consisted of data Topics to be discussed in detail in this paper include:
integration, fine-grid model development, scale-up, coarse- development of the reservoir simulation model, application of
grid model development, history matching, and the proposed methodology for probabilistic forecasting,
deterministic/probabilistic forecasting. What distinguishes the development of the QoHM variable, selection of the QoHM
probabilistic forecasting of this brown-field development from acceptance criteria, the two-step Monte Carlo simulation
that of comparable green-field developments is the available approach, final project results, and lessons learned / best
production history and the need to retain the integrity of the practices in probabilistic forecasting of brown-field
history match while fully assessing the impact of uncertainty development projects.
levels in key geologic and project variables (uncertain
parameters). Introduction
The methodology employed in this study is similar to the The Nemba field is located in Blocks 67, 68, 79, and 80 of
framework proposed by Castellini et. al.1, but was adapted to Area B in the Block 0 Concession of the Republic of Angola
use more off-the-shelf technologies. This methodology uses in 350 to 420 ft (105 to 128 m) water depths (please see
both single variable input modifications (sensitivity analysis) Figure 1). The field was discovered in 1990 with the drilling
2 SPE 95869

of Well 67-01X. The productive interval of the Nemba field is developments because the entire range of uncertainty in the
the prolific Pinda formation which consists of mixed uncertain parameters can be applied with little or no constraint
carbonate/siliclastic reservoir rock deposited in a foreshore, imposed by historical production data (possibly constrained by
barrier island complex. The accumulation is contained within DST results, extended well test results, EPS results, or analog
a highly faulted anticline containing two separate four-way dip analysis). For brown-field development cases, however, the
enclosed structures filled to the spillpoint. The reservoir is a availability of historical production data, the desire to retain
saturated oil reservoir containing an initial 870 ft (265 m) gas the integrity of a history matched simulation model, and the
column and 515 ft (157 m) oil column. need to assess the full impact of the uncertainty range must all
Procurement and construction of the South Nemba be considered.
platform began in December 1995. Pre-drilling of the South For brown-field development cases the use of Monte Carlo
Nemba development wells began in 1996. In addition to the simulation and decision tree analysis has also been discussed
South Nemba platform, a floating EPS concept was evaluated in the literature. Murtha4 described the use of Monte Carlo
in 1994. The Nemba EPS was planned to allow for production simulation in conjunction with volumetric analysis, empirical
through three subsea wells prior to the installation of the South correlations, and exponential decline for estimating
Nemba platform. The objectives of the EPS were to improve probabilistic reserves; Guyaguler and Horne18 described the
the economics of the development through early oil production use of decision trees for assessing uncertainty in well
and to gather dynamic reservoir data to aid in finalizing the placement in history matched models; and Wongnapapisan24
design of the South Nemba platform and for front-end described the use of decision tree analysis to optimize field
engineering of a follow-up North Nemba platform. The performance for a wide range of reservoir management
Nemba EPS was sanctioned in January 1995 and production decisions. The integration of geological uncertainties into this
through the EPS began in January 1996. latter decision risk analysis was not discussed.
In June 1998, production from the Nemba EPS was The use of multiple geological scenario and realizations
replaced with the commissioning of the South Nemba platform has also been discussed.10,16,20,27 in the context of brown-field
consisting of twelve production wells (nine new wells and the developments. In this approach, multiple geological scenarios
three tied-back EPS wells) and two gas injection wells. In (depositional systems, degree of diagenesis, degree of faulting,
2000, three new infill wells were drilled from the South etc.) or multiple realizations within one geological scenario
Nemba platform. In March 2001, the North Nemba platform are generated to identify P10, P50, and P90 geological models
was commissioned and consisted of nine production wells and for use in history matching and forward predictions. This
three gas injection wells. methodology has the disadvantage of the need to construct
From the outset, the Nemba field was planned as a multiple geological models, to screen these models to identify
minimum flare facility. Produced gas was to be re-injected the three models of interest, and to perform history matches on
into the gas cap to avoid flares and conserve reservoir energy; these models.
providing injection support to crestal and flank production. The use of experimental design for brown-field
The gas injection capacity was designed for the anticipated development cases is also problematic as it provides no a
production rates, however; reservoir voidage from oil, water, priori assurance that the simulated and historical production
and fuel-gas withdrawals has resulted in an instantaneous data will agree. To address this concern a recent study25 used
VRR of approximately 0.85. A VRR less than unity, in an iterative experimental design approach, in which the range
addition to the pre-injection EPS production, have resulted in of uncertainty of the input parameters was narrowed during
significant pressure decline in the reservoir. each iteration until all simulation cases in the selected design
In order to arrest the pressure decline, a new compression honored the historical production data. A possible
train is being evaluated for potential use in the Nemba field. disadvantage of this approach is that if the history match is
The objectives of the new compression are to utilize make-up honored for all experimental design cases, then the final
gas from adjacent fields to increase the Nemba VRR to values uncertainty ranges will be limited by one or two cases (i.e.,
greater than or equal to unity, to store make-up gas for future combinations of all pessimistic or all optimistic uncertainty
use in an Angolan Liquefied Natural Gas (LNG) plant, and to levels) which may represent very unlikely cases (< P1.).
use excess compression capacity to optimize production once Therefore, this approach may not adequately characterize
the make-up gas supply is diverted to the LNG plant. To reservoir performance in cases with mixed pessimistic and
support the evaluation of the Nemba compressor upgrade, optimistic uncertainty levels applied across the original
reservoir characterization and simulation studies were uncertainty ranges.
performed in 2003/2004. An alternative approach for brown-field situations was
The application of statistical methods to analyze historical proposed by Castellini et. al.1, in which the original
production performance to aid in forward forecasting has been uncertainty ranges were retained in the experimental design
discussed in the literature in the 1950-1960’s.2,3 During the and response surface modeling, but a two-step Monte Carlo
1990’s there has been an increased interest in probabilistic approach was used to filter input data to remove those that
reservoir analysis.1,4-27 For green-field development cases, could potentially degraded the history match quality. It was
probabilistic analysis techniques are generally based on the this approach1 that was used in the Nemba study. The
use of Monte Carlo simulation,5,6,8,12,13 decision tree analysis,13 objective of this paper, therefore, is to discuss the details of
multiple geological scenarios or realizations,15 experimental how this approach was applied to the probabilistic assessment
design,4,7,11-15,17,19,21,23,26 or combinations of these and decision analysis of the brown-field, Nemba compressor
approaches15,18. These techniques are adequate for green-field upgrade project.
SPE 95869 3

Deterministic Reservoir Simulation ii. Visually inspect all history match results to
The deterministic modeling of the Nemba field consisted of a identify those resulting in acceptable and
conventional reservoir simulation and forecasting study.28-30 unacceptable matches
While multiple realizations of the geologic model were iii. Define Quality of History Match (QoHM)
generated, upscaled using dynamic upscaling methods31, and variable(s) to “quantify” the quality of the
reviewed with scale-up diagnostics; only a single realization match
was considered for history matching and forward predictions. iv. Determine the acceptance criteria for the QoHM
The coarse-grid model resulting from the selected realization variable(s) that describe acceptable history
was a 58x129x33 structured grid (246,906 total cells) match results
containing 136,675 active cells and 33,794 non-neighbor v. Develop response surface(s) for the QoHM
connections. The non-neighbor connections were required to variable(s)
properly model flow across mapped faults (approximately 80 vi. Develop response surfaces for the project result
seismically mapped faults) and a geologic unconformity. variables (EUR, discounted incremental oil
Figure 2 shows the simulation grid used in the study. recovery, etc.)
Two fluid descriptions were use in the deterministic d. Perform a two-step Monte Carlo simulation:
simulation, a three-component (three hydrocarbon i. Obtain inputs for a Monte Carlo trial from all
components) and a six-component EOS description. To uncertain parameter CDFs
expedite the history match, the majority of the simulation ii. Estimate the QoHM variables from the trial
effort was performed with the three-component EOS model. inputs and the QoHM response surface
The six-component model was updated with history match iii. If the QoHM estimate indicates a feasible trial,
modifications on an approximate one to two week basis and (i.e., acceptable history match), then determine
was run to ensure alignment with the three-component model. the project results from the trial inputs and
Figure 3 shows the history match quality for the two appropriate response surfaces. Otherwise
deterministic EOS models at the field level. Comparable discard the unfeasible trial inputs.
matches were achieved at the well level. In addition to the rate iv. Continue the Monte Carlo simulation for the
and static pressure data matched at the field and well levels, desired number of trials
rate and pressure data were also matched at the geologic v. Develop the CDFs (S-Curves) for the project
sequence level through PLT and SFT matching, respectively. from all valid trials
Figure 4 and Figure 5 show examples of the PLT and SFT 3. Perform project decision analysis:
matches from the Nemba model. a. Identify representative deterministic simulation
Early in the evaluation, deterministic prediction cases were models which yield P10, P50, and P90 results
run to identify the base case model, the range of compressor b. Integrate the sub-surface uncertainty analysis into
sizes, and the scope of the project to carry forward into the the total project decision analysis using decision
probabilistic analysis. For the different compressor sizes tree analysis
analyzed, the required numbers and locations of new gas
injection wells were investigated, and the appropriate tubing Uncertain Parameter Identification (Step 1)
size/completions that allowed for the full utilization of the At the beginning of the analysis, a comprehensive uncertainty
compressor were determined. Later in the evaluation, management assessment was performed to identify the key
deterministic, representative P10, P50, and P90 reservoir geologic and project design uncertainties. These uncertainties
simulation models were identified from the Monte Carlo included:
analysis and were used as an aid to integrate the sub-surface
uncertainty analysis into the total project decision analysis, to • Geologic uncertainties
develop risk mitigation strategies, and to develop upside o OGOC depth
capture strategies. o OOWC depth
o Initial water saturations, Swi
Total Project Probabilistic and Decision Analysis o Cross-fault transmissibility
The probabilistic analysis of this brown-field, upgrade project o Vertical fault-plane transmissibility
used the following methodology: o Reservoir kv/kh ratio
o Gas-oil relative permeability
1. Identify uncertain parameters o Oil-water relative permeability
2. Perform sub-surface uncertainty analysis: o Residual oil saturation to gas, Sorg
a. Validate uncertainty ranges of uncertain parameters • Drilling and completion uncertainties
with sensitivity analysis (single parameter o Completion efficiency/skin (new gas injectors only)
adjustment per simulation case) o Drilling CAPEX
b. Perform parameter screening with lower order (two- • Facilities uncertainties
level) experimental design o Compressor efficiency/downtime
c. Perform response surface modeling with higher o New facilities CAPEX
order (three-level) experimental design: o New compressor start-up date
i. Run all experimental design cases (historical, o Current facilities downtime/lost production
“do-nothing”, and “do-nothing plus project”)
4 SPE 95869

• Contractual and fiscal uncertainties to the experimental design phase (Step 2b); will generally
o Make-up gas volumes result in a wider range of uncertainty in the uncertain input
o Make-up gas compositions parameters. This is because multiple, one-sided parameter
o Oil prices adjustments either to the low or high sides in the experimental
o Incremental project OPEX design will not occur during sensitivity analysis. The use of
o LNG plant start-up date (a separate project that this wider uncertainty range during later phases did, however,
impacts the make-up gas forecasts for Nemba) result in experimental design cases with unacceptable history
matches. Those combinations of input data resulting in
In this list, the italicized bullet items were evaluated using unacceptable history matches (unfeasible input data) were
the reservoir simulation model during the sub-surface identified and excluded during the Monte Carlo simulations
uncertainty analysis in Step 2 of the above procedure; while using the QoHM filter which will be described later. It is this
the non-italicized bullet items were incorporated into the filtering process that forms the basis of the framework
overall project decision analysis through simulation post- proposed by Castellini et. al.1.
processing and decision tree analysis. It should be noted that narrowing the uncertainty range of
The preliminary uncertainty ranges for the uncertain individual uncertain parameters during the sensitivity analysis
parameters were obtained by: may still result in an undesirable reduction of the feasible
parameter space. This is because an extreme value from the
• Geologic parameters were determined from the data original uncertainty range which resulted in a poor history
acquisition program at Nemba match may still be included in the feasible parameter space
• Drilling and completion parameters were obtained from when compensated by perturbations of other uncertain
a concession-wide drilling database parameters. This being said, any adjustments to the uncertainty
• Facilities parameters (excluding compressor efficiency) range of individual uncertain parameters during sensitivity
were obtained from an industry-wide benchmarking analysis as opposed to experimental design analysis, will
study of comparable brown-field projects result in broader uncertainty ranges because honoring a history
• Compressor efficiency was obtained from the historical match with multiple, one-sided extremes (either high or low)
performance of the existing Nemba compressors is not a issue.
While every effort was made to balance the trade-off
The uncertainty in the make-up gas availability ranged between maintaining the widest possible uncertainty ranges
from a low-end estimate based on the P10 gas forecasts from and retaining the history match quality for all of the sensitivity
existing and fully sanctioned new field developments to a cases, at the end of this step of the procedure the pessimistic
high-end estimate based on an unlimited gas supply requiring kv/kh did result in a marginally unacceptable history match.
future new field discoveries to fully utilize the compressor This poor history match, however, proved to be beneficial to
capacity. The incorporation of the make-up gas composition the project as it provided a blind test for the QoHM filter.
into the uncertainty analysis is work in progress and the results The results of the sensitivity analysis are shown in Figure 6
are not reported in this paper. During the Monte Carlo through Figure 8. Since this is an incremental project, all
simulation, triangular probability density functions (PDFs), results were generated by running a “do-nothing” case and a
which are characterized by low, mode, and high estimates32, “do-nothing plus project” incremental case. Fig. 6 and Fig. 8
were used for a majority of the uncertain parameters. For show the results of the simulation cases; while Fig. 7 shows
clarity, in the remainder of this paper the term “most-likely the derived results for the incremental project. For the “do-
value” will refer to a value used in the deterministic, history nothing” case in Fig. 6, the project related uncertain
matched model while the term “mode” will refer to the parameters; make-up gas volumes, compressor efficiency, and
statistical mode value. new well completion efficiency/skin; have no impact on the
results because the project was not implemented. In the “do-
Sub-Surface Uncertainty Analysis (Step 2) nothing” case, the initial water saturation was the dominant
Sensitivity Analysis (Step 2a). The objectives of the geological uncertainty, however; Fig. 7 indicates that when the
sensitivity analysis (single parameter adjustment per project is implemented, the uncertainty in the make-up gas
simulation case) were to validate the uncertainty ranges for the supply dominates all other geologic and project uncertainties.
uncertain parameter input, to determine the impact of these
uncertainties on the project, and to act as blind tests for the Parameter Screening (Step 2b). The objective of the
detailed response surface modeling. parameter screening was to identify the main uncertain
At this step in the procedure, balancing the trade-off parameters to continue to analyze during the detailed response
between retaining the history match quality and maintaining surface modeling and Monte Carlo simulation phases of the
the widest range of input data uncertainty was an important study. While the sensitivity analysis described above resulted
consideration. To achieve this balance, the criterion used for in the broadest possible uncertainty ranges for the uncertain
the uncertainty range validation was to attempt to ensure that parameters, the results from this analysis were not deemed
the application of the extremes of any one, individual adequate for parameter screening since only one parameter per
uncertain parameter during the sensitivity analysis did not simulation case was adjusted, and the effects of parameter
result in an unacceptable history match. Performing this interactions could not be evaluated. For parameter screening
validation during the sensitivity analysis (Step 2a), as opposed the two-level, Plackett-Burman33 experimental design was
used. The use of the Plackett-Burman33 experimental design
SPE 95869 5

for parameter screening has been discussed extensively in the space. This is required in the proposed procedure because the
literature17,23, and its use will not be elaborated upon here. validation of the uncertainty ranges was performed during the
The results from the parameter screening are shown in sensitivity analysis and there is no guarantee that all of the
Figure 9. The results from the Plackett-Burman33 analysis experimental design cases will yield acceptable history
confirmed those of the sensitivity analysis, that the project is matches. The visual inspection process is a subjective process,
dominated by the make-up gas availability. The results for and for the Nemba project two team members independently
other uncertain parameters are shown in Fig. 9. This figure is reviewed each history match case to determine the quality of
in overall agreement with the comparable results shown in Fig. the match. In most cases the two team members were aligned
8. For brevity, only the “do-nothing plus project” results are in their decisions regarding the quality of the history matches.
reported here. Figure 10 shows examples of acceptable and unacceptable
During the sensitivity analysis the depths to the original history match results. The results of the visual inspection of all
contacts, OGOC and OOWC, were found to have minimal of the D-Optimal and sensitivity analysis cases are reported in
impact on the project and were excluded from the Plackett- Table 1.
Burman33 analysis. Fig. 8 is in full agreement with Fig. 9, This table shows that in this study 27 of the 40 D-Optimal
with the exception that make-up gas availability and cross- cases resulted in acceptable matches (from visual inspection).
fault transmissibility have change places in the ranking order. At this point there were two approaches for running the
This could be due to either the exclusion of the contact depths prediction cases: run only those prediction cases resulting
from the analysis or the inclusion of the parameter interactions from successful history matches or run all prediction cases.
inherent to experimental design. While it took additional simulation effort, we chose to proceed
Due to software limitations only six parameters could be with the latter approach. There were two reasons for this
analyzed with the more detailed three-level experimental decision.
design. To incorporate as many parameters as possible into the The first reason for running all of the predictions cases was
final analysis, the make-up gas volumes were combined with because the predictions from the unsuccessful matches
the compressor efficiency by applying the pessimistic volume represent valid model responses to unfeasible input data and,
forecast with the pessimistic efficiency, the most likely consequently, still contain useful information for the project.
volume forecast with the most likely efficiency, and the This is shown conceptually in Figure 11. In this figure, the
optimistic volume forecast with the optimistic efficiency. In feasible parameter space is illustrated in the center of the plot
effect, this resulted in a slightly wider uncertainty range for while the unfeasible parameter space is illustrated by the
the make-up gas uncertainty. The final uncertain parameters shaded areas. In this figure, the plot points represent the results
carried forward into the analysis were then: the make-up gas from the experimental design simulation cases. If response
availability, initial water saturation, the residual oil saturation surfaces are developed solely on successful history matches
to gas, kv/kh, relative permeability, and cross-fault (feasible input data represented by solid plot points), then
transmissibility. extrapolation of the response surface is required beyond the
last set of simulated feasible input data to the boundary of the
Response Surface Modeling (Step 2c). The objective of the feasible parameter space. If, however, the response surfaces
response surface modeling was to develop a simulation proxy are developed from all prediction cases, then this border area
that could be used in Monte Carlo simulation. It is at this point of feasible space can be determined by interpolation. The
that our procedure deviates slightly from the original second reason for performing all prediction cases is to
framework outlined by Castellini et. al.1. In the original work1, preserve the mathematical integrity of the original
a space-filling experimental design1,26 and a kriged response experimental design (for example, the objective of the D-
surface were utilized in the analysis. In order to use in-house Optimal design is to maximize the trace of the factor co-
software, however, we employed a D-Optimal experimental variance matrix23).
design and a quadratic polynomial for the response surface Figure 12 and Figure 13 show the results from all of the
models. forward prediction cases. These figures show the expected
Simulation of the Higher Order Experimental Design. behavior that the exclusion of simulation cases based on
The objective of the higher order experimental design was to unacceptable history matches narrows the range of outcomes
provide the control points for the response surface modeling. in the forward predictions. This is because the unacceptable
The D-Optimal experimental design for the six uncertain history matched experimental design cases tend to be based on
parameters identified during the parameter screening step multiple, one-sided extremes in the uncertain input data.
required 40 simulation cases for the project evaluation in Interestingly however, the prediction results from several
addition to a center point case (base deterministic case). As cases based on unacceptable history matches fell well within
discussed earlier, the Nemba compressor upgrade is an the range of outcomes from cases based on acceptable history
incremental project and requires all simulation cases to be run matches. Fig. 12 also shows that one “do-nothing” case did not
in history mode, as well as, “do-nothing” and “do-nothing plus converge. This and its corresponding “do-nothing plus
project” prediction modes. There will be additional discussion project” case were the only cases excluded from further
of this point in the next section. analysis since no project results could be discerned from these
Visual Inspection of the History Match Results. The cases.
objective of the visually inspection process was to identify Definition of QoHM Variables and Acceptance Criteria.
those history match cases that were unacceptable and, The objective of the QoHM analysis was to define variable(s)
therefore, indirectly define the limits of the feasible parameter which are readily available from the simulation results and
6 SPE 95869

which can be used to discriminate between the acceptable and A second method that can be used to reduce the
unacceptable history matches (as determined from the visual subjectivity of the proposed procedure would be to start with
inspection). For this project, two QoHM variables were used: sufficiently large acceptance criteria, so that initially no cases
a weighted sum of square errors in the cumulative oil, gas, and are identified as unacceptable. The acceptance criteria can then
water production at the end of history (Equation 1) and the be systematically reduced until the results stabilize (within the
relative error of the GOR at the end of history (Equation 2). changes expected from Monte Carlo simulation). This
approach would require no additional reservoir simulation or
= ω E +ω E +ω E . . . . (1) response surface modeling because, as discussed earlier, these
2 2 2
QoHM Cum o o g g w w
steps were performed using the entire suite of history match
cases, both acceptable and unacceptable. Effectively, this
would iteratively reduce the feasible parameter space as in
QoHM = . . . . (2)
GOR E GOR Ref. 25, but the iterations would be performed with Monte
Carlo simulation which requires seconds to run, as opposed to
Where EY is the relative error of property Y: reservoir simulation which may require days to run. In
addition the feasible parameter space then becomes bounded
− Y sim by the input data that no longer change the final results, as
= Y obs . . . . (3)
E Y opposed to the last set of experimental design simulation cases
Y obs
that were run. This approach was not attempted during this
Nemba compressor evaluation, however; it will be
In Eq. 1, the weights were used to improve the predictive investigated during the next application of the proposed
capability of the QoHMCum variable and are based on procedure.
knowledge gained during the history match. The final values It should be noted that the definitions for the QoHM
of the weights used for this project were: ωo=175.0, ωg=11.7, variables and acceptance criteria will vary from project to
and ωw=1.0. The high weight assigned to the oil phase was project. In the Nemba study, the QoHM variables were
used to ensure that the reservoir could meet its historical oil evaluated at the field level. There may be cases in which key
production; while the intermediate weight assigned to the gas wells or well patterns may be indicators of a successful history
phase was used to reflect the importance of gas production match. In these cases the QoHM variables can be developed
during the history match. It should be noted that during the and applied at the well or pattern level.
final saturation match stage of the history match the oil rates Development of Response Surfaces As stated earlier, the
were specified.28-30,34 Therefore, the simulation and observed objective of the response surface modeling was to develop
cumulative oil production should be identical and any proxies for the reservoir simulation model. In this study,
deviation from the specified oil rate was an indication of a conventional quadratic polynomial proxies to the simulator
history match problem. The intent of the high weight assigned response were generated which include the primary effects,
to the oil phase term in Eq. 1 was to identify any history match the interaction effects, and the non-linear quadratic effects.
case that could not achieve its specified rate (i.e., filter out These polynomials had the form4:
cases like those, shown in the extreme, in Fig. 10c). Finally,
R = a + ∑ b x + ∑ (∑ c x x )+ ∑ d x
the low weight applied to the water phase reflects the low 2
. . . . (4)
j ; j >i
watercuts exhibited during the historical period. An alternative i i i i i, j i j i i i

to the use of a weighted QoHM definition in Eq. 1 would have

been to use three QoHM variables, one for each phase, with For the Nemba study, polynomials were fit for both QoHM
three different acceptance criteria. The results from these variables (Eq. 1 and Eq. 2), the estimated ultimate recovery,
definitions are presented in Table 1. and the discounted incremental oil recovery (discounted at 10
The acceptance criteria used in this study were values of percent per annum). For the QoHM variables, the make-up gas
QoHMCum < 3 and QoHMGOR < 25 percent. While a relative uncertainty was excluded from the regression because it has
error of 25 percent in the GOR appears to be very high for a no impact on the quality of the history match. This would be
measure of history match quality, it must be emphasized that true for any uncertain parameter related to an incremental
this particular definition applies to a single point in time (the project.
end of the historical period). As an example, Figure 10b shows At this step in the procedure, the results from the
a history match case, D-Optimal Case 26, which has a 24.75 sensitivity analysis can be re-used. These results can be
percent error in GOR at the end of the historical period but utilized either to supplement the three-level experimental
was designated as marginally acceptable during the visual design results during the regression of Eq. 4 or as blind tests to
inspection. An early effort to incorporate the GOR trend into the regression analysis developed solely from the experimental
the analysis was attempted by developing a third QoHM design. The sensitivity analysis cases represent orthogonal
variable based on the difference in the GOR at the end of the “experiments” in any experimental design, however; a
historical period and six months prior to this time. This majority of the uncertain parameters are set to their most-
additional variable was not used in the final analysis, however, likely, history matched values. Therefore, use of the sensitivity
because its use did not alter the final results and, Eq. 1 and Eq. analysis results in the regression of Eq. 4 may result in
2 were found to do an adequate job in differentiating between anchoring the response surfaces to the deterministic history
acceptable and unacceptable history matches (see Table 1). match results. In the Nemba study, we chose to re-use the
sensitivity analysis results as blind tests to the regression
SPE 95869 7

analysis. It should be noted that the results from the two-level experimental design approach. It would be interesting to
Plackett-Burman33 design cannot be re-used exactly either as compare the final results from these different methods to
supplemental regression data or as blind tests to the response determine the relative merits and differences associated with
surface modeling. This is because these experiments have each. This, however, was not possible within the
some uncertain parameters (those excluded during the business/project cycle-time of the evaluation.
parameter screening process) that are set at either their
pessimistic or optimistic values and are inconsistent with the Project Decision Analysis (Step 3)
three-level experimental design. Representative P10, P50, and P90 Simulation Models
Table 1 shows the polynomial proxy results for the QoHM (Step 3a). The objectives of the representative, deterministic
analysis. In this table, the shaded entries indicate mis-matches P10, P50, and P90 simulation models were to aid in the
between the actual reservoir simulation (with visual integration of the sub-surface uncertainty analysis into the
inspection) results and the QoHM results. Due to the total project decision analysis, aid in the development of risk
approximate nature of the polynomial proxy equations, some mitigation strategies, and aid in the development of upside
cases that were accepted in the actual simulation analysis were capture strategies. During this step of the procedure,
not accepted in the proxy analysis (as in D-Optimal Case 14) deterministic reservoir simulation cases were identified that
and visa versa (as in D-Optimal Case 20). mimicked the P10, P50, and P90 results from Monte Carlo
Of particular note in this table is the kv/kh sensitivity case. simulation.
As discussed earlier, the pessimistic kv/kh sensitivity case Due to the time constraints for the project analysis, it
resulted in a marginally unacceptable history match. Table 1 would have been difficult, if not impossible, to perform the
shows that this case was successfully identified by the QoHM probabilistic analysis described in this paper for every variant
variables and acceptance criteria described in this paper. It of each facilities design concept considered in the project. To
should be emphasized that these sensitivity cases were run as integrate the sub-surface analysis into the overall project
true blind tests and the acceptance criteria were not adjusted decision analysis, representative deterministic models were
post-priori to “predict” this result. Overall, Table 1 indicates used. It should be emphasized that the use of representative
that the QoHM variables and acceptance criteria do a P10, P50, and P90 simulation models is an approximation
reasonable job of identifying input data combinations that because there is no guarantee that the interactions in sub-
resulted in the unacceptable history matches. In addition to the surface uncertain parameters will be identical for different
blind tests for the QoHM variables, the sensitivity cases were processes, decisions, or design concepts. Unfortunately, there
also used successfully as blind tests for the proxy equations for are no alternative approaches available that could have been
the project result variables. used within the scheduled timeframe.
In the Nemba project, the main criteria used to identify the
Two-Step Monte Carlo Simulation (Step 2d). The objectives representative simulation models were the project EUR and
of the two-step Monte Carlo simulation were to predict the discounted cumulative oil. Once the P10, P50, and P90 estimates
quality of the history match for a given set of random inputs, for these two criteria were determined from the Monte Carlo
filter out unfeasible input parameter combinations, and simulation, deterministic models were designed to yield
develop the sub-surface S-Curves (CDFs). comparable results.
During the Monte Carlo simulations, the initial water To validate the representative models, polynomial proxy
saturations and make-up gas availability were sampled using equations from the D-Optimal results were fit for the year-end
discrete PDFs; while all other uncertain parameters used cumulative oil production for all years from 2004 through
triangular PDFs. In the first step of the Monte Carlo 2030 (i.e., year-end cumulative oil produced from first-oil to
simulation, the trial inputs were used to evaluate the QoHM 2004, 2005, 2006, etc.) and, P10, P50, and P90 estimates of these
proxy equations to predict the quality of the history match that values were made. These results are the Monte Carlo results
would have been obtained with these data had they been run plotted in Fig. 12 and Fig. 13. Figure 16 and Figure 17
through reservoir simulation. If both QoHM proxies passed compare the results from the Monte Carlo derived year-end
the acceptance criteria, then the trial inputs were used to production forecasts to the deterministic simulation derived
evaluate the proxy equations for the project results. This forecasts from the representative models. It should be noted
process was repeated until the desired number of successful that if these results matched exactly, then both the EUR and
trials was achieved. For the Nemba study, this QoHM filtering discounted incremental cumulative oil criteria would also have
process typically discarded 25-30 percent of the Monte Carlo matched exactly.
trials.
The resulting S-Curves for the “do-nothing” and “do- Integration of the Sub-Surface Uncertainty Analysis with
nothing plus project” cases are shown in Figure 14 and Figure the Total Project Decision Analysis (Step 3b). The objective
15. For comparison, the S-Curves developed from a of the integration process was to combine the sub-surface
conventional, single-step, unfiltered Monte Carlo simulation results with other project related uncertainties and decisions to
are also plotted in these figures. Fig. 14 and Fig. 15 illustrate compare and select from alternate project designs. For
the same results as those in Fig. 12 and Fig. 13: that the effect comparing and selecting between alternatives, the tool-of-
of filtering out unacceptable history match cases tends to choice is decision tree analysis.13 This is because the expected
narrow the range of forward prediction outcomes. Not plotted value of the different alternatives is a direct outcome of the
in these figures are the results from multiple geological process.
scenarios/realizations or the results from an iterative
8 SPE 95869

For the decision tree analysis, the representative, significant number, if not all, of the experimental
deterministic P10, P50, and P90 simulation models were used to design cases result in acceptable history matches.
evaluate all branches of the decision tree stemming from a 3. The modification of the uncertainty range of uncertain
decision node. In the decision tree, the uncertain nodes were parameters should be performed judiciously. History
obtained from the uncertainty management assessment matching can only be used as a guide for reducing any
discussed earlier (non-italicized bullet items in the earlier uncertainty range and cannot, by itself, be used to
bullet list); while the decision nodes were based on the determine or justify these ranges. This is because an
facilities design concepts considered in the project. These extreme value in one uncertain parameter modified
design concepts included the compressor size, optimal use of alone that results in an unacceptable history match may
available deck space versus the need for a small service still be included in the feasible parameter space when
platform, construction/installation options requiring a single or combined with interactions from other uncertain
multiple mobilizations/demobilizations, etc. parameters. It is the total set of input values, not one
Figure 18 shows the total project S-Curves for a decision extreme value, which defines the feasible parameter
on compressor size that were developed with decision tree space.
analysis. In addition to the construction of the S-Curves the 4. The proposed technique can be used to estimate the
representative, deterministic simulation models were used to uncertainty range of individual uncertain parameters.
develop risk mitigation and up-side capture strategies when The results from the successful QoHM filtering can be
appropriate. In Fig. 18 the integration of the simulation results parsed to determine the minimum and maximum values
for a risk mitigation strategy applied to one of the alternatives of each uncertain parameter contained in the sampled
is also shown. feasible parameter space. It should be emphasized that
in some cases, the interactions of these data may be
Summary and Conclusions more important than the extrema of these data.
A detailed case study of the probabilistic analysis of a brown- 5. Validating/modifying the uncertainty range of the
field, incremental project has been presented for the Nemba uncertain parameters can be performed during the
field. The methodology used in this study was based on the sensitivity analysis (single parameter adjustment per
framework proposed by Castellini et. al.1. The primary simulation case). This will result in a wider range in the
objective of this framework is to minimize any reductions to uncertain inputs but does not guarantee that all
the uncertainty ranges of the uncertain parameters required to experimental design (multiple parameter adjustments
preserve the integrity of a history match. The integrity of the per simulation case) history matches will be acceptable.
history match is preserved during the Monte Carlo simulation 6. Validating/modifying the uncertainty range during
step of the process and not the experimental design step. experimental design will generally result in a narrower
While it may appear that the application of this proposed range of uncertain inputs if the criteria for modification
procedure will result in significant overhead, this is not the are to ensure that all experimental design history
case. Many of the steps discussed in this paper are required for matches are accepted. This is because most
other methods of brown-field development analysis. For experimental designs contain case(s) with multiple,
example, depending on the complexity of the project the one-sided input combinations.
sensitivity analysis, parameter screening (low order 7. There is value/knowledge in the prediction cases that
experimental design), experimental design history match cases result from poor or unacceptable history matches. This
(higher order experimental design), experimental design well is because these cases represent valid model responses
calibrations28-30, experimental design “do-nothing” predictions, to unfeasible input data. The responses from these cases
and experimental design “do-nothing plus project” predictions help to constrain the response surfaces in the vicinity of
may all be necessary for other incremental project analysis the boundary of the feasible parameter space.
methods. (Note: for brown-field, probabilistic reserve 8. The range in the outcomes of forward predictions is
estimates from an exiting development only one set of forward narrowed by honoring the historical data. This result is
predictions is required.) In contrast to these required activities, to be expected. This narrowing will occur if the
the new elements of the proposed procedure including, the historical data are honored during either the
QoHM modeling, the response surface modeling, and two-step experimental design step or the Monte Carlo simulation
Monte Carlo simulation, are the least labor intensive and time step (as demonstrated in this paper).
consuming steps in the process. It is apparent that, if a more
compact experimental design is used in the analysis, then the Acknowledgments
relative contribution from the overhead will become more We thank the managements of the member companies of the
significant. Block 0 Association (Sonangol, CABGOC, Total Petroleum
From the material presented in this paper, the following Angola Limited, and ENI Angola Production B.V.) for their
conclusions were drawn: permission to present this work. We also thank the following
1. The original framework proposed by Castellini et. al.1 corporate colleagues for their valuable contributions and
can be adapted with more conventional experimental suggestions: Jon Sheffield, Alan Bernath, Jorge Landa, Paul
designs and response surface models to allow for use of Maslanka, Stan Franklin, Steve Johnson, Steve Gross, and
off-the-shelf technologies and software. Duke Snyder. Finally, we thank our CABGOC and Block 0
2. The validation and modification of the uncertainty Association colleagues, too numerous to mention by name, for
range of uncertain parameters is inevitable to ensure a their many valuable discussions, insights, and contributions.
SPE 95869 9

Nomenclature 14. White, C.D., Willis, B., J., Narayanan, K., and Dutton, S. P.:
Roman “Identifying Controls on Reservoir Behavior Using Designed
a = Regression coefficient Simulations,” paper SPE 62971 presented at the 2000 SPE
b = Regression coefficient Annual Technical Conference and Exhibition, Dallas, TX, Oct.
1-4, 2000.
c = Regression coefficient 15. van Elk, J. F., Guerrera, L., Vijayan, K., and Gupta, R.:
d = Regression coefficient “Improved Uncertainty Management in Field Development
E = Relative error, dimensionless Studies through the Application of the Experimental Design
P = Probability level Method to the Multiple Realisations Approach,” paper SPE
QoHM = Quality of history match variable 64462 presented at the 2000 SPE Annual Technical Conference
R = Response and Exhibition, Dallas, TX, Oct. 1-4, 2000.
x = Uncertain parameter 16. Mezghani, E. M., Zabalza-Mezghani, I., and Roggero, F. R.:
“Combination of Experimental Design and Joint Modeling
Methods for Quantifying the Risk Associated With
Greek
Deterministic and Stochastic Uncertainties - An Integrated Test
ω = Weighting factor Study,” paper SPE 71620 presented at the 2001 SPE Annual
Technical Conference and Exhibition, New Orleans, LA, Sept.,
References 30 - Oct. 3, 2001.
1. Castellini, A., Landa, J. L., and Kikani, J.: ”Practical Methods 17. Friedmann, F., Chawathe, A., and Larue, D.: “Assessing
for Uncertainty Assessment of Flow Predictions for Reservoirs Uncertainty of Channalized Reservoirs Using Experimental
with Significant History – A Field Case Study,” presented at the Designs,” paper SPE 71622 presented at the 2001 SPE Annual
9th European Conference on the Mathematics of Oil Recovery, Technical Conference and Exhibition, New Orleans, LA, Sept..
Cannes, France, 30 Aug.-2 Sept., 2004. 30 - Oct. 3, 2001.
2. Chatas, A. T. and Tankie, W. W. Jr.: “Application of Statistics 18. Guyaguler, B. and Horne, R. N.: “Uncertainty Assessment of
to Production Decline Data,” Trans., AIME (1958) 399. Well Placement Optimization,” paper SPE 71625 presented at
3. Warren, J. E.: “The Performance of Heterogeneous Reservoirs,” the 2001 SPE Annual Technical Conference and Exhibition,
paper SPE 964 presented at the 39th Annual Fall Meeting of the New Orleans, LA, Sept., 30 - Oct. 3, 2001.
Society of Petroleum Engineers, Houston, TX, Oct. 11-14, 1964. 19. White, C.D., Willis, B., J., Narayanan, K., and Dutton, S. P.:
4. Damsleth, E., Hage, A. and Volden, R.: “Maximum Information “Identifying and Estimating Significant Geologic Parameters
at Minimum Cost: A North Sea Field Development Study with With Experimental Design,” SPEJ (Sept. 2001) 311.
an Experimental Design,” JPT (Dec. 1992) 1350. 20. Mishra, S., Choudhary, M. K., and Datta-Gupta, A.: “A Novel
5. Ovreberg, O., Damsleth, E., and Halderson, H. H.: “Putting Approach for Reservoir Forecasting Under Uncertainty,”
Error Bars on Reservoir Engineering Forecasts,” JPT (June SPEREE (Feb. 2002) 42.
1992) 732. 21. Kabir, C. S., Chawathe, A., Jenkins, S. D., Olayomi, A. J.,
6. Murtha, J. A.: “Incorporating Historical Data in Monte Carlo Aigbe, C. and Faparusi, D. B.: “Developing New Fields Using
Simulation,” paper SPE 26245 presented at the SPE Petroleum Probabilistic Reservoir Forecasting,” paper SPE 77564
Computer Conference, New Orleans, LA, July 11-14, 1993. presented at the 2002 SPE Annual Technical Conference and
7. Faidi, S. A., Ponting, D. K., and Eagling, T. L.: “Experimental Exhibition, San Antonio, TX, Sept. 29 – Oct. 2, 2002.
Design in Interactive Reservoir Simulation,” paper SPE 36005 22. Verbruggen, R., Pannett, S. and Stone, G.: “Understanding
presented at the SPE Petroleum Computer Conference, Dallas, Reserves Uncertainties in a Mature Field by Reservoir
TX, June 2-5, 1996. Modelling,” paper SPE 77896 presented at the 2002 SPE Asia
8. Murtha, J. A.: “Monte Carlo Simulation: Its Status and Future,” Pacific Oil and Gas Conference and Exhibition, Melbourne,
paper SPE 37932 presented at the 1997 SPE Annual Technical Australia, Oct. 8-20, 2002.
Conference and Exhibition, San Antonio, TX, Oct. 5-8, 1997. 23. White, C. D. and Royer, S. A.: “Experimental Design as a
9. Soendenaa, E. and Halderson, H. H.: “E&P; Old and New Framework for Reservoir Studies,” paper 79676 presented at the
Forecasting Issues,” paper SPE 49029 presented at the 1998 SPE SPE Reservoir Simulation Symposium, Houston, TX, Feb. 3-5,
Annual Technical Conference and Exhibition, New Orleans, LA, 2003.
Sept. 27-30, 1998. 24. Wongnapapisan, B., Boyd, F., and Hassan, Z.: “ Optimising
10. Twartz, S. K., Gorjy, F., and Milne, I. G.: “A Multiple Brown Field Redevelopment Options Using a Decision Risk
Realisation Approach to Managing Uncertainty In The North Assessment: Case Study – Bokor Field, Malaysia,” paper SPE
Rankin Gas Condensate Field, Western Australia,” paper SPE 87047 presented at the 2004 SPE Asia Pacific Conference on
50078 presented at the 1998 SPE Asia Pacific Oil and Gas Integrated Modeling for Asset Management,” Kuala Lumpur,
Conference and Exhibition, Perth, Australia, Oct. 12-14, 1998. Malaysia, March. 29-30, 2004.
11. Venkataraman, R.: “Application of the Method of Experimental 25. Peake, W. T., Abadah, M., and Skander, L.: “Uncertain
Design to Quantify Uncertainty in Production Profiles,” paper Assessment Using Experimental Design: Minagish Oolite
SPE 59422 presented at the 2000 SPE Asia Pacific Conference Reservoir,” paper SPE 91820 presented at the 2005 SPE
on Integrated Modeling for Asset Management,” Yokohama, Reservoir Simulation Symposium, Houston, TX, Jan. 31 – Feb.
Japan, April. 25-26, 2000. 2, 2005.
12. Chewaroungroaj, J., Varela, O. J., and Lake, L. W.: “An 26. Yenten, B., Castellini, A., Guyaguler, B. and Chen, W., H.: ”A
Evaluation of Procedures to Estimate Uncertainty in Comparison Study on Experimental Design and Response
Hydrocarbon Recovery Predictions,” paper SPE 59449 Surface Methodologies,” paper SPE 93347 presented at the 2005
presented at the 2000 SPE Asia Pacific Conference on Integrated SPE Reservoir Simulation Symposium, Houston, TX, Jan. 31 –
Modeling for Asset Management,” Yokohama, Japan, April. 25- Feb. 2, 2005.
26, 2000. 27. Wences, A. and Tomas, M.: “Risk Analysis in Determination of
13. Murtha, J. A.: “Risk Analysis for the Oil Industry,” Hart’s E&P the Best Exploitation Strategy Using a Calibrated Reservoir-
(Aug. 2000) Supplement. Simulation Model,” paper SPE 94804 presented at the SPE Latin
10 SPE 95869

American and Caribbean Petroleum Conference, Rio de Janeiro, 31. Durlofsky, L. J.: “Numerical Calculation of Equivalent Grid
Brazil June 20-23, 2005. Permeability Tensors for Heterogeneous Porous Media,” Water
28. Maddox, C. C. and Dalton, R. L.: Reservoir Simulation, Resources Research 27, No. 5 (May 1991) 699.
Monograph Series 13, the Society of Petroleum Engineers, 32. Murtha, J. A.: “Sums and Products of Distributions: Rules of
Richardson, TX, (1990). Thumb and Applications,” paper SPE 77422 presented at the
29. Ertekin, T., Abou-Kassem, J. H., and King, G. R.: Basic Applied 2002 SPE Annual Technical Conference and Exhibition, San
Reservoir Simulation, Textbook Series 7, the Society of Antonio, TX, Sept. 29 – Oct. 2., 2002.
Petroleum Engineers, Richardson, TX, (2001). 33. Plackett, R. L. and Burman, J. P.: “The Design of Optimal
30. King, G. R.: ”Practical Aspects of a Contemporary Reservoir Multifactorial Experiments,” Biometrika 33 No. 4 University
Simulation Study,” presented at the Seventh International Forum Press, Cambridge, U. K. (June 1946) 305.
on Reservoir Simulation, Baden-Baden, Germany, June 23 – 27, 34. Williams, M. A., Keating, J. F., and Barghouty, M. F.: “The
2003. Stratigraphic Method: A Structured Approach to History
Matching Complex Simulation Models,” SPERE (April 1998)
169.

Table 1
History Match Results from Simulation (with Visual Inspection) and from QoHM Filtering.
Case Description Used in Regression Reservoir Simulation / Visual Inpection Analysis QoHM Proxy Analysis
Analyses ? Oil Match Gas Match Water Match Overall Assessment Overall Assessment
Center Point Deterministic Base Case No Acceptable Acceptable Acceptable Pass Pass
D-Optimal 1 Experimental Design Case Yes Unacceptable Unacceptable Unacceptable Fail Fail
D-Optimal 2 Experimental Design Case Yes Unacceptable Unacceptable Unacceptable Fail Fail
D-Optimal 3 Experimental Design Case Yes Unacceptable Unacceptable Unacceptable Fail Fail
D-Optimal 4 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 5 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Pass
D-Optimal 6 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Pass
D-Optimal 7 Experimental Design Case Yes Acceptable Unacceptable Marginally Acceptable Fail Fail
D-Optimal 8 Experimental Design Case Yes Acceptable Unacceptable Acceptable Fail Fail
D-Optimal 9 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Pass
D-Optimal 10 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 11 Experimental Design Case Yes Unacceptable Unacceptable Unacceptable Fail Fail
D-Optimal 12 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 13 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 14 Experimental Design Case Yes Acceptable Marginally Acceptable Marginally Acceptable Pass Fail
D-Optimal 15 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 16 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 17 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 18 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Pass
D-Optimal 19 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 20 Experimental Design Case Yes Acceptable Marginally Unacceptable Unacceptable Fail Pass
D-Optimal 21 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 22 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 23 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 24 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 25 Experimental Design Case Yes Acceptable Unacceptable Marginally Acceptable Fail Fail
D-Optimal 26 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Fail
D-Optimal 27 Experimental Design Case Yes Marginally Acceptable Acceptable Marginally Acceptable Pass Pass
D-Optimal 28 Experimental Design Case Yes Acceptable Unacceptable Acceptable Fail Fail
D-Optimal 29 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 30 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Pass
D-Optimal 31 Experimental Design Case Yes Unacceptable Unacceptable Unacceptable Fail Fail
D-Optimal 32 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Pass
D-Optimal 33 Experimental Design Case Yes Marginally Acceptable Unacceptable Unacceptable Fail Fail
D-Optimal 34 Experimental Design Case Yes Acceptable Acceptable Acceptable Pass Pass
D-Optimal 35 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Fail
D-Optimal 36 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Fail
D-Optimal 37 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Pass
D-Optimal 38 Experimental Design Case Yes Acceptable Unacceptable Acceptable Fail Fail
D-Optimal 39 Experimental Design Case Yes Acceptable Marginally Acceptable Acceptable Pass Fail
D-Optimal 40 Experimental Design Case Yes Acceptable Unacceptable Acceptable Fail Fail
Sensitivity 1 Pessimistic Swi No Acceptable Acceptable Acceptable Pass Pass
Sensitivity 2 Optimistic Swi No Acceptable Acceptable Acceptable Pass Pass
Sensitivity 3 Pessimistic kv /k h No Acceptable Marginally Unacceptable Acceptable Fail Fail
Sensitivity 4 Optimistic kv /k h No Acceptable Acceptable Acceptable Pass Pass
Sensitivity 5 Pessimistic Cross-Fault Transmissibity No Acceptable Acceptable Acceptable Pass Pass
Sensitivity 6 Optimisitic Cross-Fault Transmissibity No Acceptable Acceptable Acceptable Pass Pass
Sensitivity 7 Pessimistic Relative Permeability No Acceptable Acceptable Acceptable Pass Pass
Sensitivity 8 Optimisitic Relative Permeability No Acceptable Acceptable Acceptable Pass Pass
Sensitivity 9 Pessimistic Sorg No Acceptable Acceptable Acceptable Pass Pass
Sensitivity 10 Optimistic Sorg No Acceptable Acceptable Acceptable Pass Pass
SPE 95869 11

History Match
Produced Watercut
Full-Field Model Matches

0.5
Historical 3 Component EOS 6 Component EOS
(A) Watercut Match
0.4

Produced Watercut (Fraction)

0.3

0.2

0.1

0.0
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
10
Historical 3 Component EOS 6 Component EOS
(B) GOR Match

Produced GOR (MSCF/STB)

6

0
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
125
Historical 3 Component EOS 6 Component EOS
(C) Oil Rate Match (Note: Oil Rate is

100
Oil Production Rate (MBOPD)

75
Figure 1 Nemba Field in the Block 0 Concession
50

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58
1
2
3
4
5 25
6
7
8
9
10 Fault 10
11
12
13 B4
14 0
15
16
17
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
18 Fault 9
19 B10 Time (Year)
20
Fault 13
21 B5 B5 Fault 12
22 B5
23
24
25
26
27
28
Figure 3 History Match Quality of the Deterministic Three-
29
30
31
32
Fault 14
B1 B1 B1
Component and Six-Component EOS Models at the Field Level
33
34
35 B9
36 B9
37
38
39
40
41
42 Fault 8
43
44
45 Fault 18
46 Fault 6 B11B11
47 B3 B11
48 B3
49 Fault 17
50
51
52 B8
Fault 4
53 B8 B6
54 Fault 15 B6
55
56 A16
57 A16
58 A16
59
60
61 Fault 3
Fault 5
62 Fault 2 A12
63
64 Fault 7 A17
65 A17
66
67
68 A15A15
69 A15 A10 A3
70 A10
71 A10
72
73
74
75 A1
76 A2
77
78
79
80 Fault 1 A8 A8 A7
81 A6
82
83 A5
84
85
86
87
88 A13A13
89 A13
90 Fault 16a
91
92 A9
93
94
95 A4 A4
96
97
98
99
100
101 A14
102 A14
103 Fault 16b
104
105
106 A11
107 A11
108 A11
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129

Figure 2 Nemba Field Reservoir Simulation Grid

12 SPE 95869

Injector 1 Injector 2
21-MAY-03 18-SEP-02 Producer 4 Producer 5
7-Jun-00 29-Oct-01
Cumulative Gas %
Cumulative Gas % 8600 8600
0 20 40 60 80 100 Measure SFT Measured SFT
8500 0 20 40 60 80 100 Simulated SFT Simulated SFT
8200 Initial Pressure
Measured PLT Measured PLT Initial Pressure
8800 8800
Simulated PLT Simulated PLT
8600

8400 9000 9000

8700

8800 9200 9200

8600

Depth (ft TVD s

Depth (ft TVD s

8900
9400 9400
Depth

Depth (ft TVD s

9000 8800
9600 9600

9100

9800 9800
9000
9200

10000 10000
9300
9200

9400 10200 10200

9500 9400
10400 10400
4500 4750 5000 5250 5500 4500 4750 5000 5250 5500
Producer 1 Producer 2 Pressure (psia)
22-OCT-98 31-JAN-00 Pressure (psia)

Cumulative Liquid %
Cumulative Liquid % Producer 1 Producer 3
0 50 100
0 50 100 9480
20-Nov-96 31-May-97
9400 Measured PLT
Measured PLT 8800
Simulated PLT
Simulated PLT 8600
Measured SFT Measured SFT
9450 9500
Simulated SFT Simulated SFT
9000
Initial Pressue
Initial Pressure
8800
9500
9520
9200
9000
9550
9540
9400
Depth (ft TVD s

9200
Depth (ft TVD s

9600
Depth ((ft TVD s

9560
Depth (ft TVD s

9600
9400
9650

9580
9800
9600
9700

9600
10000 9800
9750

9620 10200 10000

9800

9850 9640 10400 10200

Figure 4 Examples of Nemba Well PLT Matches

10400
10600
5000 5200 5400 5600
5200 5400 5600 5800
Pressure (psia) Pressure (psia)

Figure 5 Examples of Nemba Well SFT Matches

SPE 95869 13

Pareto Diagram Pareto Diagram

Sensitivity Analysis Two-Level Plackett-Burman Analysis
"Do Nothing" Cases "Do-Nothing plus Project" Cases
Analysis Variable: Cumulative Oil at 2030 Analysis Variable: Cumulative Oil at 2030

Swi 6.69 Swi 6.50

Sorg 6.43
Sorg 5.57
Kv/Kh 5.20
Kv/Kh 3.93
Rel. Perm. 4.71

Rel. Perm. 3.77

Cross-Fault Trans. 1.91

Fault-Plane Trans. 1.02 Cross-Fault Trans. 3.45

OOWC 0.01
Make-Up Gas 2.21

OGOC 0.01
Fault-Plane Trans. 2.02
Make-Up Gas 0.00
Compressor Eff. 1.10
Compressor Eff. 0.00

Completion Eff. 0.00 Completion Eff. 0.10

0 1 2 3 4 5 6 7 8 9 10 0 1 2 3 4 5 6 7 8 9 10
Effects Estimate (Absolute Value) Effects Estimate (Absolute Value)

Figure 6 Pareto Diagram of Uncertainties for “Do-Nothing” Cases Figure 9 Pareto Diagram of Uncertainties for “Do-Nothing plus
33
during Sensitivity Analysis Project” Cases during Plackett-Burman Analysis

D-Optimal Analysis
Visual Inspection Analysis
Full-Field Results

0.5
Historical D-Optimal Case 1 D-Optimal Case 20 D-Optimal Case 26
Pareto Diagram
Sensitivity Analysis (A) Watercut Match
Incremental Project Cases 0.4
Analysis Variable: Cumulative Oil at 2030 D-Optimal Case 1: Unacceptable
D-Optimal Case 20: Unacceptable
Produced Watercut (Fraction) D-Optimal Case 26: Acceptable
Make-Up Gas 18.43
0.3
Cross-Fault Trans. 2.61

Kv/Kh 2.17
0.2
Sorg 1.78

Compressor Eff. 1.61

0.1
Rel. Perm. 1.15

Fault-Plane Trans. 0.98

0.0
Swi 0.85 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005

Completion Eff. 0.01 20

Historical D-Optimal Case 1 D-Optimal Case 20 D-Optimal Case 26
OOWC 0.01 (B) GOR Match

OGOC 0.01
16
D-Optimal Case 1: Unacceptable
0 2 4 6 8 10 12 14 16 18 20 D-Optimal Case 20: Unacceptable
Effects Estimate (Absolute Value) D-Optimal Case 26: Marginally Acceptable
Producing GOR (MSCF/STB)

Figure 7 Pareto Diagram of Uncertainties for Project Incremental 12

Results during Sensitivity Analysis

8

0
1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Pareto Diagram
Sensitivity Analysis 125
"Do-Nothing plus Project" Cases Historical D-Optimal Case 1 D-Optimal Case 20 D-Optimal Case 26
Analysis Variable: Cumulative Oil at 2030
(C) Oil Rate Match (Note: Oil Rate is Specified)

Swi 6.01 100

D-Optimal Case 1: Unacceptable
D-Optimal Case 20: Unacceptable
Sorg 5.41
D-Optimal Case 26: Acceptable

Kv/Kh 4.27
Oil Rate (MBOPD)

75
Rel. Perm. 3.99

Make-Up Gas 2.61

50
Cross-Fault Trans. 1.31

Fault-Plane Trans. 1.04

Compressor Eff. 0.23

25

Completion Eff. 0.01

OOWC 0.01
0
OGOC 0.01 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005
Time (Year)
0 1 2 3 4 5 6 7 8 9 10
Effects Estimate (Absolute Value)

Figure 8 Pareto Diagram of Uncertainties for “Do-Nothing plus Figure 10 Examples of History Matches from the D-Optimal
Project” Cases during Sensitivity Analysis Analysis
14 SPE 95869

Limit of Acceptable History Match

Sub-Surface S-Curve (CDF)
“Do-Nothing” Case
Response “R”

Cumulative Probability
Limit of Acceptable History Match

Feasible Input Parameter Space

Input Parameter Set “x”

Acceptable Unacceptable Interpolated Simulator Extrapolated
Simulator History Simulator History Proxy (Response Simulator
Match Results Match Results Surface) Proxy

Figure 11 Conceptual Justification for Performing Response

Surface Modeling on all Simulation Cases

Cumulative Produced Oil: "Do-Nothing" Cases

All D-Optimal Cases (Valid and Invalid History Matches)
600

500
Cumulative Produced Oil (MMSTB)

Estimated Ultimate Recovery

400
With QoHM Filter Without QoHM Filter
300

Figure 14 Cumulative Distribution Functions for the “Do-Nothing”

200
Case from Sub-Surface Probabilistic Analysis
100

0
1995 2000 2005 2010 2015 2020 2025 2030
Sub-Surface S-Curve (CDF)
Year “Do-Nothing plus Project” Case
Historical Data Case with Valid History Match Case with Invalid History Match
Monte Carlo P10 Monte Carlo P50 Monte Carlo P90

Figure 12 “Do-Nothing” Predictions from all D-Optimal

Simulation Cases (Note: Non-Convergence of One Simulation
Cumulative Probability

Case)

Cumulative Produced Oil: "Do-Nothing plus Project" Cases

All D-Optimal Cases (Valid and Invalid History Matches)
600

500
Cumulative Produced Oil (MMSTB)

400

300

200

100
Estimated Ultimate Recovery

0 With QoHM Filter Without QoHM Filter

1995 2000 2005 2010 2015 2020 2025 2030
Year
Historical Data Case with Valid History Match Case with Invalid History Match Figure 15 Cumulative Distribution Functions for the “Do-Nothing
Monte Carlo P10 Monte Carlo P50 Monte Carlo P90
plus Project” Case from Sub-Surface Probabilistic Analysis

Figure 13 “Do-Nothing plus Project” Predictions from all

D-Optimal Simulation Cases
SPE 95869 15

Cumulative Produced Oil: "Do-Nothing" Cases

Comparison of P10, P50, and P90 Models 100

600 90

80
500
Cumulative Produced Oil (MMSTB)

70

Cumulative Probability (percent)

400
60

300 50

40
200

30

100
20

0 10
1995 2000 2005 2010 2015 2020 2025 2030
Year
0
0
Historical Monte Carlo P10 Monte Carlo P50 Monte Carlo P90 NPV ($MM)
Deterministic P10 Deterministic P50 Deterministic P90
Compressor Size 1 Compressor Size 2 Compressor Size 3 (Original) Compressor Size 3 (with Risk Mitigation)

Figure 16 Comparison of Monte Carlo Generated P10, P50, and P90 Figure 18 Total Project Decision from Decision Tree Analysis
Forecasts and Representative P10, P50, and P90 Simulation Model Integrating Sub-Surface, Drilling, Facilities, and Fiscal/Contractual
Forecasts for the “Do-Nothing” Case Uncertainties (Note the application of a risk mitigation strategy for
Compressor Size 3 from representative deterministic models)

Cumulative Produced Oil: "Do-Nothing plus Project" Cases

Comparison of P10, P50, and P90 Models

600

500
Cumulative Produced Oil (MMSTB)

400

300

200

100

0
1995 2000 2005 2010 2015 2020 2025 2030
Year

Historical Monte Carlo P10 Monte Carlo P50 Monte Carlo P90
Deterministic P10 Deterministic P50 Deterministic P90

Figure 17 Comparison of Monte Carlo Generated P10, P50, and P90

Forecasts and Representative P10, P50, and P90 Simulation Model
Forecasts for the “Do-Nothing plus Project” Case